Colonoscopy is recommended for colorectal cancer screening as well as the diagnosis and treatment of various gastrointestinal diseases. It is one of the most common ambulatory care procedures, with estimates of over 14 million procedures in 20002. It is also technically complex, and colonoscopy quality and outcomes may be quite variable. Given its widespread application and high costs, understanding the quality and outcomes of colonoscopy as applied in routine clinical practice is vital. Use of administrative data to measure health care quality is attractive, but the validity of administrative data for routine measures that might indicate quality is not established. This proposal provides important methodological first steps towards using administrative data to measure colonoscopy quality. In our first specific aim, we will obtain colonoscopy reports from medical records of a diverse group of providers to use to determine the accuracy of administrative claims for colonoscopy. These reports will be linked to colonoscopy claims in the Medicare claims database. We will examine the sensitivity, specificity, positive predictive value, and negative predictive value of the claims for a variety of important colonoscopy quality indicators, including incomplete colonoscopy, detection rates for colon polyps or cancers, and use of biopsy or polypectomy. Developing an algorithm to determine colonoscopy indication from administrative data is another important methological step towards using such data in quality measurement. For example, the rate of polyp or cancer detection may be different in patients undergoing this procedure for colorectal cancer screening, as compared to patients who clinical signs or symptoms. The ability to examine quality measures in administrative data is therefore dependent upon understanding colonoscopy outcomes. In our second and third specific aims, we will develop and test an algorithm to determine colonoscopy indications from claims data. In an initial, training subset of our database, we will examine the diagnostic and procedure codes associated with the colonoscopy claim as well as codes from prior claims to find individual codes or patterns of coding that predict colonoscopy outcome. The various algorithms we develop will be tested in a second, validation subset of our database to find the algorithm with optimal performance characteristics.